UK Invests $273M in Super AI Computer, Key Takeaways at UK 2023 AI Safety Summit; Microsoft Advances Small AI Models


The UK government is investing $273 million in building Isambard-AI, one of the world’s most powerful supercomputers, equipped with thousands of NVIDIA Superchips and expected to begin operations in summer 2024 at the University of Bristol. 

Isambard-AI, constructed by Hewlett Packard Enterprise, will use 5,448 NVIDIA GH200 Grace Hopper Superchips to achieve over 21 exaflops of AI performance, significantly surpassing the fastest UK supercomputer. 

This announcement was made during the UK’s AI Safety Summit, which aimed to establish global AI governance rules and featured significant developments, including:

  1. The Bletchley Declaration: 28 countries, including the US, UK, and China, signed the Bletchley Declaration to promote transparency and scientific collaboration in AI safety practices.
  2. Kamala Harris Addresses AI Harms: US Vice President Kamala Harris highlighted AI-related issues such as discrimination and misinformation and announced the Biden administration’s commitment to managing AI’s societal risks.
  3. Elon Musk’s AI Concerns: Elon Musk expressed concerns about advanced AI’s potential existential dangers, emphasizing the need for responsible development and guidance.
  4. AI Development Halt Consideration: Mustafa Suleyman, co-founder of DeepMind, suggested a possible temporary halt in AI development soon but clarified that current AI models like GPT-4 do not pose significant harm.
  5. UK’s Investment in Isambard-AI: The UK government’s £225 million investment in Isambard-AI, aimed at advancing AI in healthcare, energy, and climate modelling, is part of the country’s efforts to lead in AI alongside allies like the US.
  6. Global AI Dominance Competition: The summit revealed a high-stakes competition among major players like the US, EU, and China to set AI rules and standards, although cooperation is also emphasized to prevent unchecked AI development. 

Additionally, Microsoft‘s Phi 1.5 model has become multimodal, capable of interpreting images and providing a cost-effective alternative to large AI models like GPT-4. This development could help democratize AI technology and alleviate shortages in graphics processors for tasks like ChatGPT.